Systematic Identi cation of Sca olds Representing Di erent Types of Structure-Activity Relationships

In medicinal chemistry, it is of central importance to understand structureactivity relationships (SARs) of small bioactive compounds. Typically, SARs are analyzed on a case-by-case basis for sets of compounds active against a given target. However, the increasing amount of compound activity data that is becoming available allows SARs to be explored on a large-scale. Moreover, molecular sca olds derived from bioactive compounds are also of high interest for SAR analysis. In general, sca olds are obtained by removing all substituents from rings and from linkers between rings. This thesis aims at systematically mining compounds for which activity annotations are available and investigating relationships between chemical structure and biological activities at the level of active compounds, in particular, molecular sca olds. Therefore, data mining approaches are designed to identify sca olds with di erent structural and/or activity characteristics. Initially, sca old distributions in compounds at di erent stages of pharmaceutical development are analyzed. Sets of sca olds that overlap between di erent stages or preferentially occur at certain stages are identi ed. Furthermore, a systematic selectivity pro le analysis of public domain active compounds is carried out. Sca olds that yield compounds selective for communities of closely related targets and represent compounds selective only for one particular target over others are identi ed. In addition, the degree of promiscuity of sca olds is thoroughly examined. Eighty-three sca olds covering 33 chemotypes correspond to compounds active against at least three di erent target families and thus are considered to be promiscuous. Moreover, by integrating pairwise scaffold similarity and compound potency di erences, the propensity of sca olds to form multi-target activity or selectivity cli s and, in addition, the global sca old potential of individual targets are quantitatively assessed, respectively. Finally, structural relationships between sca olds are systematically explored. Most sca olds extracted from active compounds are found to be involved in substructure relationships and/or share topological features with others. These substructure relationships are also compared to, and combined with, hierarchical substructure relationships to facilitate activity prediction.